Page 786 - Emerging Trends and Innovations in Web-Based Applications and Technologies

International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
Each model is executed with a fixed random state of 42 to independence assumption holds reasonably true. The model
ensure consistent and reproducible results. has three variations: Multinomial, Bernoulli, and Gaussian.
The Multinomial Naïve Bayes is particularly suited for text
Figure 2 Comparison of Most Common Words in Fake vs
data, as it models word occurrences in documents. Input
Real News Articles
features are generally derived from BoW or TF-IDF, and the
4. Machine Learning Models model computes the likelihood of a document being fake or
4.1. Random Forest real based on word frequencies. Naïve Bayes is
Random Forest is a widely used ensemble machine learning computationally efficient and works well for smaller
method, particularly effective in text classification tasks like datasets. However, its independence assumption can limit its
fake news detection. It functions by building multiple ability to capture complex dependencies between words,
decision trees during training, and the final output is based especially in more nuanced tasks where contextual
on the majority vote from individual tree classifications (for relationships matter [18]. Despite these limitations, Naïve
classification) or an average prediction (for regression). For Bayes is often chosen for fake news detection due to its
Natural Language Processing (NLP), Random Forest uses speed, simplicity, and reasonable performance, particularly
features extracted from text, such as term frequency-inverse when used as a baseline model or alongside other methods.
document frequency (TFIDF) or word embeddings, to 4.4. Neural Networks
perform classification [16].
Neural Networks, particularly deep learning architectures,
A key strength of Random Forest lies in its ability to process have significantly advanced NLP by enabling models to
high-dimensional datasets, a typical characteristic of textual autonomously learn features from textual data. In fake news
data. By combining predictions from multiple trees, it detection, neural networks use preprocessed text inputs,
reduces overfitting and enhances generalization. In fake such as word embeddings or token sequences, to understand
news detection, Random Forest can detect patterns in word intricate relationships and contextual details. The simplest
usage and sentence structure across large datasets, helping form, Feedforward Neural Networks (FNNs), processes input
identify important features. Additionally, it ranks feature features via fully connected layers.
importance, allowing for the interpretation of which terms or However, more advanced architectures, such as Recurrent
phrases are most significant in predictions. However, Neural Networks (RNNs) and Long Short-Term Memory
effective hyperparameter tuning—such as the number of (LSTM) networks, are better suited for sequential data like
trees and maximum tree depth—is necessary for optimal text, capturing the flow of information across a document
results. Despite its reliance on aggregated features, which and identifying inconsistencies that may indicate fake news.
might miss some contextual nuances, Random Forest Transformer-based models, like BERT, further enhance this
remains a reliable and interpretable method for tasks like
by leveraging self-attention mechanisms to understand
fake news detection.
context and relationships across entire documents. Neural
4.2. Support Vector Machine (SVM) networks are particularly strong in processing large,
Support Vector Machine (SVM) is a powerful supervised unstructured data and uncovering deep patterns that other
learning technique well-suited for text classification tasks models might miss. However, they require significant
due to its robust handling of high-dimensional and sparse computational resources and vast labeled datasets for high
data. In the context of NLP tasks like fake news detection, accuracy. Techniques like dropout regularization are often
SVM identifies a hyperplane that best separates data points used to prevent overfitting. Despite these challenges, neural
from different classes. By maximizing the margin between networks are currently the state-of-the-art in NLP and excel
the classes, SVM ensures reliable and precise classification. It in tasks such as fake news detection.
employs kernel functions—such as linear, polynomial, or 4.5. Logistic Regression
radial basis functions (RBF)—to map data into higher- Logistic Regression is a simple but effective linear model
dimensional spaces where linear separation is achievable. commonly employed for binary classification tasks, including
For text classification, features are commonly derived from fake news detection. It predicts the probability of an instance
bag-of-words (BoW), TF-IDF, or word embeddings. SVM belonging to a particular class by applying a logistic function
excels in fake news detection by analyzing word to the weighted sum of input features. In NLP, features for
distributions and patterns [17].
Logistic Regression are often derived from techniques like
One of SVM's major advantages is its ability to manage noisy bag-of-words, TF-IDF, or word embeddings. Despite its
and imbalanced datasets, making it well-suited for real- simplicity, Logistic Regression performs remarkably well in
world fake news detection tasks. However, its computational fake news detection, especially when combined with robust
cost increases with large datasets, especially when nonlinear feature engineering. The model assigns weights to features
kernels are used. Proper tuning of hyperparameters, such as based on their relevance to the classification, making it
kernel selection and regularization parameters, is critical to interpretable. For example, it can highlight specific terms or
achieving optimal performance. Despite these challenges, phrases strongly associated with fake news. However,
SVM remains a popular tool for NLP due to its scalability and Logistic Regression struggles to capture nonlinear
efficiency in high-dimensional settings. relationships and contextual nuances in text, which are vital
4.3. Naïve Bayes for identifying more complex fake news patterns.
Regularization methods, such as L1 and L2 penalties, are
Naïve Bayes is a probabilistic approach often used in text
frequently used to avoid overfitting, especially with high-
classification tasks due to its simplicity and effectiveness.
dimensional data. While not as sophisticated as advanced
Based on Bayes' theorem, it assumes that features are
models like neural networks, Logistic Regression remains a
conditionally independent given the class label. Despite the solid baseline for fake news detection due to its simplicity,
"naïve" assumption, Naïve Bayes performs well in NLP tasks, efficiency, and interpretability.
such as fake news detection, particularly when the

IJTSRD | Special Issue on Emerging Trends and Innovations in Web-Based Applications and Technologies Page 776